Artificial Intelligence – Debt Debate pt. II

AI is giving the debt markets a bout of indigestion.

• The current situation is serving as a reminder to the AI industry that it is dangerous to fund an arms race with debt.
• The instability in the Middle East is causing investors to move into cash, which in turn is causing the cost of debt to rise, which is sending ripples of concern through the AI industry.
• Around $400bn is expected to be borrowed in the USA alone, as part of the funding for $650bn of AI data centre construction, which I think is a far riskier proposition than raising the money via equity, which never has to be paid back.
• AI is affecting debt markets in two ways:
  • First, Private Credit: which are funds where investors can invest in unlisted debt that often carries both a higher interest rate and higher risk.
  • These instruments are also illiquid, meaning that only when the bond matures, and the principal is repaid can the investor get back the money he or she put in.
  • Apollo has joined Morgan Stanley and BlackRock in limiting the amount of money that investors can pull out of their funds in a sign that many investors want to head for the exits.
  • These funds have the option to cap redemptions at 5%, which they have exercised as redemption requests are running at something closer to 10% to 12% of fund value.
  • This has been going on since before the current instability in the Middle East and is a reflection of how investors are currently viewing the software and services sector.
  • The prevailing view is that AI will put an entire sector out of business, and as many of these private credit funds have lent money to software companies, investors want out.
  • Second, Balance Sheet Degradation: which is where investors are raising debt to invest in AI companies and where the AI companies themselves are loading up their balance sheets with debt to finance the massive AI infrastructure roll-out.
  • SoftBank is the latest to hit the headlines, with its CFO admitting that SoftBank is likely to breach its stated cap of 25% of net debt to find value as it raises more debt to invest in OpenAI.
  • This follows on from Google, Amazon, Oracle and many others who are also raising debt to fund AI data centre rollouts.
  • The problem I have with this is that RFM Research has demonstrated that the business model of providing AI compute is uneconomic in its current form.
  • This means that unless something changes, many companies are going to struggle to pay back the debt that they have raised.
  • This, in turn, will trigger a liquidity crunch, which, as all of these companies are now financially interrelated, will affect everybody fairly severely.
• When one piles on the current situation with huge volatility in the cost of energy, it is not difficult to envisage how a liquidity crunch caused by the war in the Middle East could destabilise the AI industry.
• This is why I have argued for years that raising debt to fund an arms race is generally a bad idea, and the last times this happened, the Internet Bubble and the 2008 financial crisis were the painful results.
• This is why the economics of Nvidia’s latest Vera Rubin system will be crucial, as Nvidia is promising a 5x increase in revenues per GW of AI capacity under certain circumstances.
• We have heard these sorts of things before, but they have failed to deliver on better economics, as revenue per GW has been stuck at $10bn for the last 3 years.
• I think that we are still quite far from a serious crisis in AI financing, but these things always occur very slowly and then very suddenly and all at once.
• Hence, I am not tempted at all to buy SoftBank’s shares even though the shares have fallen by 50% since these rumblings began.
• Instead, I remain happy to stick with memory via Samsung, inference at the edge via Qualcomm and nuclear power.